Panel surveys, such as the Survey of Income and Program Participation and the Consumer Expenditure Survey, interview respondents every 3 or 4 months, but ask the respondents for monthly data. A typical finding in such surveys is that changes in responses to a question are relatively small for adjacent months within a reference period but much more abrupt for adjacent months across reference periods. Previous studies have attributed this “seam effect” either to underreporting of changes within the periods or to overreporting of changes across them. In the present studies, we attempt to distinguish these possibilities, using an experimental method that allows us to gauge respondents' accuracy as well as the number of times they change their answers. The studies produced seam effects and accompanying evidence for forgetting of queried information and bias toward constant responses within the reference period. In general, seam effects appear to increase as a function of the demands on memory. We also find that separating questions with the same content in the survey instrument decreases the seam effect. To account for these data, we propose a model in which respondents' answers are initially based on attempted memory retrieval. Inability to recall leads to (possibly biased) guessing or subsequent repetition of an earlier answer.